To reduce problems in comparing gains in the rate of students who progress to the next level of education, a nonlinear transformation is recommended. Two such transformations, probits and logits, are presented. Log linear models are recommended for an appropriate statistical test of sampling error. (Author/GDC)